Specify Task Type

Name Your Project

Give your project a unique name


Choose Your Task Type

Studio currently supports the following task types:

General Image Annotation: Use general image annotation for 2D image workflows. This project type supports geometry based labels with the option to specify custom attributes or metadata for each label. These geometry annotations include bounding boxes, polygons, lines, points, and cuboids - which are annotated with unique IDs.

2D Semantic Segmentation: Use 2D semantic segmentation for pixel-wise annotation which is useful to classify both the things of the world (e.g discrete objects) amorphous regions (e.g., sky and ground). We support both partial and full semantic segmentation as well as instance masks for panoptix segmentation.

Text Collection/Categorization: Use text collection/categorization as a generalized project type to support various classification tasks. These can include audio transcription, categorization, information collection and extraction, sentiment analysis, and more. There’s also the option to specify consensus if you’d like multiple responses per task.

Document Transcription: Use document transcription to extract key fields from your documents. The output includes a bounding box for each field along with a text transcription. We can also support named semantic relationship between fields like invoice amount to the amount listed.

Named Entity Recognition: Use this to tag entities found in text using specific labels. For example, you could set up a task to label any mention of time or duration in a set of text. This project type also supports named semantic relationships between entities.

Video Playback Annotation: Use video playback annotation for geometry based object tracking and event tagging. Use bounding boxes, polygons, lines, points, and cuboids to track objects and people frame to frame with unique and consistent object IDs. Create named events tied to specific frames and optionally add verbal or written descriptions for each event.

LIDAR Annotation: Use LIDAR annotation to annotate your point cloud data with 3D cuboids in Scale’s platform. While autonomous driving sequences is the primary use case we support, Scale’s platform can also ingest 3D point cloud reconstructions to support a wide range of 3D annotation projects across many use cases such as sidewalk robotics, drone delivery, AR/VR, or more.


LIDAR annotation is available to Pro and Enterprise customers only.

If you’re interested in using Studio’s LIDAR annotation capabilities, please contact the team at scale_studio@scale.com.


Choose Your Pipeline

For Text Collection/Categorization or Named Entity Recognition task types, you have the ability to specify the kind of pipeline you want to use.

Standard Pipeline: This pipeline uses 1 attempt for every task, with an additional number of reviews for each attempt. The project default is set to 0 reviews, but you can change the number of reviews if desired.

Consensus Pipeline: Choose this pipeline if you want to maximize your quality or if you have a subjective task which may have multiple right answers. Using a consensus pipeline will send each task to 3 different attempters and the majority answer will be the one that is delivered to you as the “final” response.

Freeform Pipeline: Choose this pipeline if you have freeform text, where it is more difficult to measure accuracy. In this pipeline, each task gets paired up in attempter/reviewer pairs. Unlike the standard pipeline, which is only 1 attempter per task, the freeform pipeline allows multiple attempts - and each of these attempts will also have a corresponding review. If the first attempter/reviewer pair align on an answer, then the task gets passed onto the second pair, and once they align on an answer, it gets passed on yet again - for however long you specified. This makes measuring accuracy more feasible because responses are now binary.

Updated a month ago